Methodology
How would we know? Like this.
The Visibility Score measures how present a business is in AI answers. We put a fixed panel of twelve of your buyers' real questions to Claude, with live web search on, and record whether each answer names the business, whether it names it first, and who it names instead. The score is the share of the twelve that named the business, a number between 0 and 100, measured on the same questions every time, which is the whole point: the reading after the fixes, and each monthly reading with monitoring, compares with the first.
Methodology
Four steps, and the same four every time.
The question panel
Questions phrased the way your customers really ask them, from the problem question ("where can I find...") through the comparison question to the one that names you directly. The panel is twelve questions, built once for your trade and your location and then left alone, because readings are only comparable if the questions behind them are identical.
One assistant, with the web on
Every question goes to Claude, made by Anthropic, with live web search on, in a fresh session, the way a buyer would type it. ChatGPT, Gemini and Perplexity read their own indexes and can answer differently, so we never report Claude's answer as theirs. Each answer is dated, because answers drift from one run to the next.
The evaluation
Each answer is read for three things: were you named, were you named first, and who else got named. We keep each answer verbatim, with the pages it read, so any verdict we recorded can be checked against the answer it came from, and we send you that answer when you ask. A person checks every report before it is sent.
The score
The score is the share of the panel's questions that named you, from 0 to 100: named in 3 of 12 is 25. Beside it you get the questions that named you first, the businesses named instead of you, and, with monitoring, this month against last on every question.
What each answer is read for
Named. First. Or someone else.
Named
Your name appears in the answer. This is what the score counts: named in 3 of the 12 questions is a score of 25.
First
You are the first business the answer names. It is reported beside the score, question by question, rather than folded into it, so the score stays a count anyone can check.
Named instead
Where the answer does not name you, the businesses it named are recorded. The ones named most often are the competitors the audit takes apart, page by page.
Honesty
What the score can do, and what it can't
Can
- Make readings comparable, because the panel, the assistant and the yardstick never change
- Show gaps precisely, since every lost question is a piece of work someone can pick up
- Attribute impact, by showing which measure won which questions
- Put the competition in context, by showing who gets named instead of you, and for which questions
Cannot
- Predict revenue. It measures visibility, not bookings
- Guarantee any individual answer. Assistants are probability machines, which is exactly why monitoring measures every month
- Rate your other channels. Google rankings and ads have their own yardsticks
Questions about the method
If you are going to trust a number, interrogate it first.
How is AI visibility actually measured?
By asking. A fixed panel of twelve questions your buyers really put to an assistant goes to Claude, with live web search on, each in a fresh session and typed the way a buyer would type it. Every answer is kept word for word, with the businesses it named, in order, and the pages it read. From those we record whether you were named, whether you came first, and which businesses appeared where you did not. The share of the questions that named you becomes one number between 0 and 100.
What is a visibility score?
A single number from 0 to 100: the share of your panel's questions in which the assistant named you. Named in 3 of 12 is 25. Being named first, and who was named instead, are reported beside it rather than folded in, so you can recompute the number from the answers yourself. On its own the number means little; its job is to be comparable to the last reading, which is why the panel is frozen once an engagement starts.
Does the same question get the same answer every time?
No. Assistants are probability machines, not lookup tables, and a second ask can name a different set of businesses. So no single answer is read as a verdict. Each question is asked once per reading and the reading is dated; the score is the pattern across twelve questions rather than any one of them, and monitoring compares months on the same questions instead of reacting to one answer.
How is this different from a rank tracker?
A rank tracker tells you what position a page holds in a list of links. There is no list here and no position: an assistant writes a paragraph and names two or three businesses in it, or it does not name you. So what gets measured is presence inside an answer, not an ordinal, and the competitor data is who got named instead of you rather than who outranked you.
Can I see the raw answers behind my score?
Yes. Every answer is kept verbatim with the date and the assistant that produced it, and we send the text of any answer behind your score when you ask, so any verdict in your report can be traced back to the text it came from. A score you cannot audit is a score you have to take on trust, and this is the one thing we ask you not to take on trust.
What does the score not tell me?
Revenue. It measures visibility, not bookings, and anyone selling you a straight line between the two is guessing. It also cannot guarantee an individual answer, which is why it is measured repeatedly rather than promised, and it says nothing about your other channels — Google rankings and paid campaigns have their own yardsticks and should keep them.
Run it against your own site.
The visibility check is a smaller version of this same measurement. It costs nothing and comes back in 48 hours.